Nvidia ,Rtx2080ti,2080,2070, information thread. Reviews and prices September 14.

[LTC8K6]

https://www.techradar.com/news/the-...ally-twice-as-powerful-as-the-nvidia-gtx-1080

Twice as fast as the 1080? Its twice in a few areas if you turn on DLSS.

I see 40-60% for the 4 games on the right, which seems amazing given the # of shaders and the clocks.

 

[Kenmitch]

I see 40-60% for the 4 games on the right, which seems amazing given the # of shaders and the clocks.

Were do you see the clocks?

 

[n0x1ous]

Were do you see the clocks?

he doesnt. he just means based on the known base and boost clocks not being very different

 

[Det0x]

I see 40-60% for the 4 games on the right, which seems amazing given the # of shaders and the clocks.

While its possible you are not allowed to watch videos from this guy by your contract, but i suggest to take a look and risk the hardware.

https://youtu.be/CT2o_FpNM4g?t=684

He is one of the few that actually leaked the correct specs ahead of the release.

Your looking at a 20% to 25% performance increase with the 10x0 generation compared to the 20x0 generation in the vast majority of games today and for the next few years. (non "RTX games")

*edit*

Had the wrong person quoted

You are not allow to accuse others
of being "shills".

AT Mod Usandthem

 

[Kenmitch]

he doesnt. he just means based on the known base and boost clocks not being very different

Those clocks don't really mean much as the gpus always boost higher than what's on the box.

 

[n0x1ous]

Those clocks don't really mean much as the gpus always boost higher than what's on the box.

right agreed.....curious to know about gpu boost 4.0

 

[LTC8K6]

Those clocks don't really mean much as the gpus always boost higher than what's on the box.

Why would Turing boost much higher than Pascal?

 

[rainy]

I see 40-60% for the 4 games on the right, which seems amazing given the # of shaders and the clocks.

Have you ever heard that: lies, damned lies and marketing?

 

[LTC8K6]

Have you ever heard that: lies, damned lies and marketing?

What's the point if reviewers are just going to show that it's false? Why bother? Reviews will be out in time for orders to be cancelled.

 

[StinkyPinky]

Hopefully we can see some leaks soon.

 

[Kenmitch]

Why would Turing boost much higher than Pascal?

Trick question?

Boost 4.0
Better cooling FE vs FE....Boost is temp dependent which I'm sure you already knew. As the temp goes up boost goes down.

I'm sure the slide was showing FE vs FE and not some aftermarket 1080 which maintains a higher boost.

 

[PeterScott]

While its possible you are not allowed to watch videos from this guy by your contract, but i suggest to take a look and risk the hardware.

https://youtu.be/CT2o_FpNM4g?t=684

He is one of the few that actually leaked the correct specs ahead of the release.

Your looking at a 20% to 25% performance increase with the 10x0 generation compared to the 20x0 generation in the vast majority of games today and for the next few years. (non "RTX games")

*edit*

Had the wrong person quoted

He "leaked":

RTX Titan
RTX 2080 8gb TU104 50% Faster than 1080
RTX 2070 7gb TU106 40% Faster than 1070
GTX 2060 5gb TU116 27% Faster than 1060


There was no Titan.
It's doubtful that 2070 is using TU106 (partially disabled TU104 more likely), and it's 8GB not 7GB.

IMO, it's wasn't a leak, it was just summarizing the rumors with more guess work, that was hit and miss.

 

[LTC8K6]

Trick question?

Boost 4.0
Better cooling FE vs FE....Boost is temp dependent which I'm sure you already knew. As the temp goes up boost goes down.

I'm sure the slide was showing FE vs FE and not some aftermarket 1080 which maintains a higher boost.

How much more clock can you get from better fan cooling, though? We're not talking about liquid cooling here.

Boost 4.0 might be better efficiency rather than higher boost levels. It might be related to turning off the tensor cores if they aren't needed, for example.

 

[antihelten]

Which aren't part of the actual GPU performance and are disingenous to include. The point is if the GPU is x% faster, not how the other bottlenecks in your sytem are holding the GPU back.

Of course I/O is part of GPU performance, I mean how are you going to deliver any data for your compute units to work on without I/O.

Besides even if I/O was somehow not part of GPU performance (which again makes no sense), you are missing the forest for the trees. I/O was just a single example, the point is that there are tons of elements that impact final GPU performance beyond just the pure CUDA core compote performance (which itself is also made up of many elements).

Give the leaked slide above show ~50% gain with DLSS OFF for a 2080 vs 1080, with only 15% more shader units, my 30% baseline estimate is likely on the pessimistic side for 2080Ti vs 1080Ti.

A bunch of cherry picked titles from Nvidia has nothing to do with baseline estimates.

Also a baseline estimate doesn't mean "this is where Nvidia should end up", it means "this is where Nvidia should end up at minimum". As such baseline estimates are inherently pessimistic.

With that being said, the numbers from Nvidia do indeed look nice. Here's hoping that they also hold up in independent reviews.

 

[PeterScott]

A bunch of cherry picked titles from Nvidia has nothing to do with baseline estimates.

It's funny how we have 20 pages of complaining that NVidia didn't show any non RTX benchmarks, and how they could have just shown one slide.

Now that slide with a bunch of games is released, and now we will get 20 more pages complaining that they are cherry picked, not the games you wanted to see.

Damned if they do, damned if they don't.

 

[PeterScott]

How much more clock can you get from better fan cooling, though? We're not talking about liquid cooling here.

Boost 4.0 might be better efficiency rather than higher boost levels. It might be related to turning off the tensor cores if they aren't needed, for example.

It will definitely have overkill cooling when it comes to old games not using the new systems. The cooling system will be designed to cool SM/RT/Tensor cores running in parallel. There should be a lot less heat when just running SM work loads.

 

[sze5003]

It's funny how we have 20 pages of complaining that NVidia didn't show any non RTX benchmarks, and how they could have just shown one slide.

Now that slide with a bunch of games is released, and now we will get 20 more pages complaining that they are cherry picked, not the games you wanted to see.

Damned if they do, damned if they don't.

I was kind of happy with that slide. They should have showed that at the presentation. You have to imagine they weren't benching 10 year old games as well so probably bf 1, the older tomb raider, Hitman, overwatch, probably titles from a year or two ago.

I'm curious about the Ti comparison now for sure.

 

[psolord]

By the time 2070 is released, pricing will be around the same for both cards, the 1080ti maybe a bit cheaper. Which means I'd rather have a 2070 with the extra functionality, especially since I don't game at 4k.

The big question is whether the cut down RT cores on the 2070 are good enough to actually enable RT in games.

Is it possible, that the 2070 has only the graphics cores disabled and not the RT cores? Compared to the 2080 that is?

 

[psolord]

So they are claiming 50% increase without DLSS and 100% with DLSS from 1080 to 2080?

Ok things got more interesting. A lot more.

I'm confused. DLSS is an antialiasing method right? So what exactly does it do, that it makes the game run twice as fast on the 2080?

Are they replacing the built in AA method of each respective game, with DLSS? Because then that wouldn't be apples to apples comparison, although if you can switch say 8X MSAA for DLSS and have the same result, at twice the speed, then yes that would be great. I kinda find smaa good enough though. I'm a peasant!

 

[alcoholbob]

I'm confused. DLSS is an antialiasing method right? So what exactly does it do, that it makes the game run twice as fast on the 2080?

Are they replacing the built in AA method of each respective game, with DLSS? Because then that wouldn't be apples to apples comparison, although if you can switch say 8X MSAA for DLSS and have the same result, at twice the speed, then yes that would be great. I kinda find smaa good enough though. I'm a peasant!

Presumably running AA process on RT cores instead of shader cores reduces/eliminates the performance penalty?

 

[PeterScott]

Is it possible, that the 2070 has only the graphics cores disabled and not the RT cores? Compared to the 2080 that is?

It has 6 RT Gigarays, vs 8 Gigarays fro the 1080, and significantly less RTX Ops or whatever other bizarre Ray Tracing specs exists. So it looks like they cut down RT and Tensor cores on RTX 1070.

I'm confused. DLSS is an antialiasing method right? So what exactly does it do, that it makes the game run twice as fast on the 2080?

Are they replacing the built in AA method of each respective game, with DLSS? Because then that wouldn't be apples to apples comparison, although if you can switch say 8X MSAA for DLSS and have the same result, at twice the speed, then yes that would be great. I kinda find smaa good enough though. I'm a peasant!

The must be replacing some other unspecified AA method, likely an expensive one. Which is why you have to take manufacture benchmarks with a grain of salt. They will always portray their product in a favorable way.

DLSS is actually one the main features I want to see tested for visual quality.

 

[krumme]

I can not do with a 30% increase for 2080 vs 1080 so ordered a 2080ti and cancelled my 2080.
That should do it for 4k 60Hz rock solid and 60 fps 1080 RT demos.Like the big die. Calms me.

And is pretty sure Ryan and others is right. Nv margins will take a huge dive now because that big die and expensive gddr6.
Might as well support them while they go from 63 margin to a more normal 53 next quarter.
/s

But lets wait and see reviews. I have my computer a hot place so not sure it will work out. And dont really favor fermis but always get them.....hmm

Edit: kept the 2080 order and a 2080ti and will wait for test. The 2080ti might be to much fermi. I guess its borderline. Like vega 64.

 

[Kenmitch]

How much more clock can you get from better fan cooling, though? We're not talking about liquid cooling here.

I can't think of any way to determine that without more data. Viewing the coolers the 2080 has a advantage as far as temps go it looks like. How boost 4.0 works? Stock fan profile for the 1080 I believe keeps the fan off till it hits a certain temperature. By the time the fan starts ramping up the cooler is already heat soaked so it isn't able to maintain as high of a boost. Does the 2080's cooler do the same thing?

 

[antihelten]

It's funny how we have 20 pages of complaining that NVidia didn't show any non RTX benchmarks, and how they could have just shown one slide.

Now that slide with a bunch of games is released, and now we will get 20 more pages complaining that they are cherry picked, not the games you wanted to see.

Damned if they do, damned if they don't.

Back when Nvidia launched Pascal they also had a bunch of comparisons, but they used common and popular titles (Assassins creed, GTA V, Titanfall, BF4, SW BF, Thief, RotTR, and the Witcher 3).

With Turing they are using tech demos (Infiltrator), obscure chinese MMOs (JX3), a bunch of games where Nvidia felt the need to point out that they had HDR turned on (FF XV, Hitman 2, Shadow of War, ME: Andromeda), and then a handful of more relevant games (PUBG, ARK, Shadow of the tomb raider, and Wolfenstein 2). Unfortunately of the more relevant games, two are multiplayer games, that tend to be problematic to reliably benchmark (PUBG and ARK), and one is currently unreleased and thus doesn't have final drivers for either GPU (shadow of the tomb raider).

If that's not cherry picking, then what else would you call it?

Also for what it's worth I never really cared about Nvidia not providing any benchmarks during the presentation, I did however care about the fact that they didn't provide any kind of indication of non-RTX performance. A simple "Turing is X% faster in non-RTX games" statement would have been fine with me (and they did actually provide such simple statements for the Pascal presentations).

Are they replacing the built in AA method of each respective game, with DLSS? Because then that wouldn't be apples to apples comparison, although if you can switch say 8X MSAA for DLSS and have the same result, at twice the speed, then yes that would be great. I kinda find smaa good enough though. I'm a peasant!

For the Infiltrator demo at least, the comparison appears to be between DLSS and TAA (that's what they were demoing at least).

 

[TheF34RChannel]

I see 40-60% for the 4 games on the right, which seems amazing given the # of shaders and the clocks.

You'd think they'd put this in the presentation. Regardless, I only prefer independent reviews as those of manufacturers always look heavenly and are, for better or worse, a salesman slide. I'll wait and see.

And if the Ti replaced the Titan, the 2080 the previous Ti and so on, why are they comparing the 1080 with the 2080?